aac.c

 * @param   crc flag indicating the presence of CRC checksum
 * @param   cnt length of TYPE_FIL syntactic element in bytes
 *
 * @return  Returns number of bytes consumed from the TYPE_FIL element.
 */
static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) {
    // TODO : sbr_extension implementation
    av_log_missing_feature(ac->avccontext, "SBR", 0);
    skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type
    return cnt;
}

/**
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
 *
 * @return  Returns number of bytes consumed.
 */
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc, GetBitContext * gb) {
    int i;
    int num_excl_chan = 0;

    do {
        for (i = 0; i < 7; i++)
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));

    return num_excl_chan / 7;
}

/**
 * Decode dynamic range information; reference: table 4.52.
 *
 * @param   cnt length of TYPE_FIL syntactic element in bytes
 *
 * @return  Returns number of bytes consumed.
 */
static int decode_dynamic_range(DynamicRangeControl *che_drc, GetBitContext * gb, int cnt) {
    int n = 1;
    int drc_num_bands = 1;
    int i;

    /* pce_tag_present? */
    if(get_bits1(gb)) {
        che_drc->pce_instance_tag  = get_bits(gb, 4);
        skip_bits(gb, 4); // tag_reserved_bits
        n++;
    }

    /* excluded_chns_present? */
    if(get_bits1(gb)) {
        n += decode_drc_channel_exclusions(che_drc, gb);
    }

    /* drc_bands_present? */
    if (get_bits1(gb)) {
        che_drc->band_incr            = get_bits(gb, 4);
        che_drc->interpolation_scheme = get_bits(gb, 4);
        n++;
        drc_num_bands += che_drc->band_incr;
        for (i = 0; i < drc_num_bands; i++) {
            che_drc->band_top[i] = get_bits(gb, 8);
            n++;
        }
    }

    /* prog_ref_level_present? */
    if (get_bits1(gb)) {
        che_drc->prog_ref_level = get_bits(gb, 7);
        skip_bits1(gb); // prog_ref_level_reserved_bits
        n++;
    }

    for (i = 0; i < drc_num_bands; i++) {
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
        n++;
    }

    return n;
}

/**
 * Decode extension data (incomplete); reference: table 4.51.
 *
 * @param   cnt length of TYPE_FIL syntactic element in bytes
 *
 * @return Returns number of bytes consumed
 */
static int decode_extension_payload(AACContext * ac, GetBitContext * gb, int cnt) {
    int crc_flag = 0;
    int res = cnt;
    switch (get_bits(gb, 4)) { // extension type
        case EXT_SBR_DATA_CRC:
            crc_flag++;
        case EXT_SBR_DATA:
            res = decode_sbr_extension(ac, gb, crc_flag, cnt);
            break;
        case EXT_DYNAMIC_RANGE:
            res = decode_dynamic_range(&ac->che_drc, gb, cnt);
            break;
        case EXT_FILL:
        case EXT_FILL_DATA:
        case EXT_DATA_ELEMENT:
        default:
            skip_bits_long(gb, 8*cnt - 4);
            break;
    };
    return res;
}

/**
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
 *
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
 * @param   coef    spectral coefficients
 */
static void apply_tns(float coef[1024], TemporalNoiseShaping * tns, IndividualChannelStream * ics, int decode) {
    const int mmm = FFMIN(ics->tns_max_bands,  ics->max_sfb);
    int w, filt, m, i;
    int bottom, top, order, start, end, size, inc;
    float lpc[TNS_MAX_ORDER];

    for (w = 0; w < ics->num_windows; w++) {
        bottom = ics->num_swb;
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
            top    = bottom;
            bottom = FFMAX(0, top - tns->length[w][filt]);
            order  = tns->order[w][filt];
            if (order == 0)
                continue;

            // tns_decode_coef
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);

            start = ics->swb_offset[FFMIN(bottom, mmm)];
            end   = ics->swb_offset[FFMIN(   top, mmm)];
            if ((size = end - start) <= 0)
                continue;
            if (tns->direction[w][filt]) {
                inc = -1; start = end - 1;
            } else {
                inc = 1;
            }
            start += w * 128;

            // ar filter
            for (m = 0; m < size; m++, start += inc)
                for (i = 1; i <= FFMIN(m, order); i++)
                    coef[start] -= coef[start - i*inc] * lpc[i-1];
        }
    }
}

/**
 * Conduct IMDCT and windowing.
 */
static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
    IndividualChannelStream * ics = &sce->ics;
    float * in = sce->coeffs;
    float * out = sce->ret;
    float * saved = sce->saved;
    const float * swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
    const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
    const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
    float * buf = ac->buf_mdct;
    DECLARE_ALIGNED(16, float, temp[128]);
    int i;

    // imdct
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
        if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
            av_log(ac->avccontext, AV_LOG_WARNING,
                   "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
                   "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
        for (i = 0; i < 1024; i += 128)
            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
    } else
        ff_imdct_half(&ac->mdct, buf, in);

    /* window overlapping
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
     * and long to short transitions are considered to be short to short
     * transitions. This leaves just two cases (long to long and short to short)
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
     */
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
        (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, ac->add_bias, 512);
    } else {
        for (i = 0; i < 448; i++)
            out[i] = saved[i] + ac->add_bias;

        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, ac->add_bias, 64);
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      ac->add_bias, 64);
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      ac->add_bias, 64);
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      ac->add_bias, 64);
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      ac->add_bias, 64);
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
        } else {
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, ac->add_bias, 64);
            for (i = 576; i < 1024; i++)
                out[i] = buf[i-512] + ac->add_bias;
        }
    }

    // buffer update
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
        for (i = 0; i < 64; i++)
            saved[i] = temp[64 + i] - ac->add_bias;
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
    } else { // LONG_STOP or ONLY_LONG
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
    }
}

/**
 * Apply dependent channel coupling (applied before IMDCT).
 *
 * @param   index   index into coupling gain array
 */
static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index) {
    IndividualChannelStream * ics = &cc->ch[0].ics;
    const uint16_t * offsets = ics->swb_offset;
    float * dest = sce->coeffs;
    const float * src = cc->ch[0].coeffs;
    int g, i, group, k, idx = 0;
    if(ac->m4ac.object_type == AOT_AAC_LTP) {
        av_log(ac->avccontext, AV_LOG_ERROR,
               "Dependent coupling is not supported together with LTP\n");
        return;
    }
    for (g = 0; g < ics->num_window_groups; g++) {
        for (i = 0; i < ics->max_sfb; i++, idx++) {
            if (cc->ch[0].band_type[idx] != ZERO_BT) {
                for (group = 0; group < ics->group_len[g]; group++) {
                    for (k = offsets[i]; k < offsets[i+1]; k++) {
                        // XXX dsputil-ize
                        dest[group*128+k] += cc->coup.gain[index][idx] * src[group*128+k];
                    }
                }
            }
        }
        dest += ics->group_len[g]*128;
        src  += ics->group_len[g]*128;
    }
}

/**
 * Apply independent channel coupling (applied after IMDCT).
 *
 * @param   index   index into coupling gain array
 */
static void apply_independent_coupling(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index) {
    int i;
    for (i = 0; i < 1024; i++)
        sce->ret[i] += cc->coup.gain[index][0] * (cc->ch[0].ret[i] - ac->add_bias);
}

/**
 * channel coupling transformation interface
 *
 * @param   index   index into coupling gain array
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
 */
static void apply_channel_coupling(AACContext * ac, ChannelElement * cc,
        void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * sce, ChannelElement * cc, int index))
{
    int c;
    int index = 0;
    ChannelCoupling * coup = &cc->coup;
    for (c = 0; c <= coup->num_coupled; c++) {
        if (ac->che[coup->type[c]][coup->id_select[c]]) {
            if (coup->ch_select[c] != 2) {
                apply_coupling_method(ac, &ac->che[coup->type[c]][coup->id_select[c]]->ch[0], cc, index);
                if (coup->ch_select[c] != 0)
                    index++;
            }
            if (coup->ch_select[c] != 1)
                apply_coupling_method(ac, &ac->che[coup->type[c]][coup->id_select[c]]->ch[1], cc, index++);
        } else {
            av_log(ac->avccontext, AV_LOG_ERROR,
                   "coupling target %sE[%d] not available\n",
                   coup->type[c] == TYPE_CPE ? "CP" : "SC", coup->id_select[c]);
            break;
        }
    }
}

/**
 * Convert spectral data to float samples, applying all supported tools as appropriate.
 */
static void spectral_to_sample(AACContext * ac) {
    int i, type;
    for (i = 0; i < MAX_ELEM_ID; i++) {
        for(type = 0; type < 4; type++) {
            ChannelElement *che = ac->che[type][i];
            if(che) {
                if(che->coup.coupling_point == BEFORE_TNS)
                    apply_channel_coupling(ac, che, apply_dependent_coupling);
                if(che->ch[0].tns.present)
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
                if(che->ch[1].tns.present)
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
                if(che->coup.coupling_point == BETWEEN_TNS_AND_IMDCT)
                    apply_channel_coupling(ac, che, apply_dependent_coupling);
                imdct_and_windowing(ac, &che->ch[0]);
                if(type == TYPE_CPE)
                    imdct_and_windowing(ac, &che->ch[1]);
                if(che->coup.coupling_point == AFTER_IMDCT)
                    apply_channel_coupling(ac, che, apply_independent_coupling);
            }
        }
    }
}

static int aac_decode_frame(AVCodecContext * avccontext, void * data, int * data_size, const uint8_t * buf, int buf_size) {
    AACContext * ac = avccontext->priv_data;
    GetBitContext gb;
    enum RawDataBlockType elem_type;
    int err, elem_id, data_size_tmp;

    init_get_bits(&gb, buf, buf_size*8);

    // parse
    while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
        elem_id = get_bits(&gb, 4);
        err = -1;

        if(elem_type == TYPE_SCE && elem_id == 1 &&
                !ac->che[TYPE_SCE][elem_id] && ac->che[TYPE_LFE][0]) {
            /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
               instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
               encountered such a stream, transfer the LFE[0] element to SCE[1] */
            ac->che[TYPE_SCE][elem_id] = ac->che[TYPE_LFE][0];
            ac->che[TYPE_LFE][0] = NULL;
        }
        if(elem_type < TYPE_DSE) {
            if(!ac->che[elem_type][elem_id])
                return -1;
            if(elem_type != TYPE_CCE)
                ac->che[elem_type][elem_id]->coup.coupling_point = 4;
        }

        switch (elem_type) {

        case TYPE_SCE:
            err = decode_ics(ac, &ac->che[TYPE_SCE][elem_id]->ch[0], &gb, 0, 0);
            break;

        case TYPE_CPE:
            err = decode_cpe(ac, &gb, elem_id);
            break;

        case TYPE_CCE:
            err = decode_cce(ac, &gb, ac->che[TYPE_CCE][elem_id]);
            break;

        case TYPE_LFE:
            err = decode_ics(ac, &ac->che[TYPE_LFE][elem_id]->ch[0], &gb, 0, 0);
            break;

        case TYPE_DSE:
            skip_data_stream_element(&gb);
            err = 0;
            break;

        case TYPE_PCE:
        {
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
            if((err = decode_pce(ac, new_che_pos, &gb)))
                break;
            err = output_configure(ac, ac->che_pos, new_che_pos);
            break;
        }

        case TYPE_FIL:
            if (elem_id == 15)
                elem_id += get_bits(&gb, 8) - 1;
            while (elem_id > 0)
                elem_id -= decode_extension_payload(ac, &gb, elem_id);
            err = 0; /* FIXME */
            break;

        default:
            err = -1; /* should not happen, but keeps compiler happy */
            break;
        }

        if(err)
            return err;
    }

    spectral_to_sample(ac);

    if (!ac->is_saved) {
        ac->is_saved = 1;
        *data_size = 0;
        return buf_size;
    }

    data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
    if(*data_size < data_size_tmp) {
        av_log(avccontext, AV_LOG_ERROR,
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
               *data_size, data_size_tmp);
        return -1;
    }
    *data_size = data_size_tmp;

    ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);

    return buf_size;
}

static av_cold int aac_decode_close(AVCodecContext * avccontext) {
    AACContext * ac = avccontext->priv_data;
    int i, type;

    for (i = 0; i < MAX_ELEM_ID; i++) {
        for(type = 0; type < 4; type++)
            av_freep(&ac->che[type][i]);
    }

    ff_mdct_end(&ac->mdct);
    ff_mdct_end(&ac->mdct_small);
    return 0 ;
}

AVCodec aac_decoder = {
    "aac",
    CODEC_TYPE_AUDIO,
    CODEC_ID_AAC,
    sizeof(AACContext),
    aac_decode_init,
    NULL,
    aac_decode_close,
    aac_decode_frame,
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
    .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
};